Development and Internal Validation of a Predictive Maintenance Management System for Improving the Reliability of Central Cooling and Standalone Air-Conditioning Systems: A Case Study of the Bank of Tanzania, Dar es Salaam

Reliable operation of central cooling and standalone air-conditioning systems is essential to environmental stability, equipment protection and operational continuity in mission-critical facilities such as a central bank. Maintenance of these systems in Tanzania, however, remains predominantly reactive and time-based, producing frequent failures, excessive downtime, and avoidable costs. This study developed and internally validated a Predictive Maintenance Management System (PMMS) to improve the reliability of the central cooling and standalone air-conditioning systems at the Bank of Tanzania (BoT), Dar es Salaam. A mixed-methods case-study design was adopted, with the facility as the unit of inference. A census of the 35 maintenance personnel was conducted (33 usable responses; 94.3%), and a separate equipment-level data set of 51 complete failure-event records (of 52 over 2023–2025) was compiled from maintenance logs and condition-monitoring sensors. Data were analysed using descriptive statistics, Pareto analysis, failure mode and effects analysis (FMEA), Pearson correlation, stepwise multiple linear regression, and machine learning (Random Forest and Support Vector Machine). Compressor-related and refrigerant-leak modes dominated the failure burden; four modes accounted for 63.8% of downtime, and five exceeded the FMEA action threshold (RPN ≥ 200). The regression model explained 78% of the variance in time to failure (R² = 0.78; F(3,47) = 55.78; p < 0.001), with vibration, bearing temperature, and current-draw deviation as significant predictors. In internal validation, Random Forest achieved 87.5% classification accuracy (AUC-ROC 0.92) in development and 75% in an eight-unit field pilot. On this basis, the PMMS is projected to raise system availability from 98.9% to 99.6% and to deliver net annual savings of about TZS 48.5 million with a seven-month payback. The study demonstrates that integrating IoT-enabled condition monitoring with statistical and machine-learning techniques provides a technically feasible, statistically validated, and economically viable PMMS for institutional cooling systems in a developing-economy setting.